Difference in fine-tuned models linked in README and used in interact.py script

The pre-trained and fine-tuned model, given in the README of the repo under the 'Pretrained model' heading (linked [here](https://s3.amazonaws.com/models.huggingface.co/transfer-learning-chatbot/finetuned_chatbot_gpt.tar.gz)) is different from what gets used when we run `python interact.py`.
In the script utils.py, the link for the fine_tuned model has been given in the script as [https://s3.amazonaws.com/models.huggingface.co/transfer-learning-chatbot/gpt_personachat_cache.tar.gz](https://s3.amazonaws.com/models.huggingface.co/transfer-learning-chatbot/gpt_personachat_cache.tar.gz).

Are these two different models because when I tried to run the earlier fine-tuned model which was linked with the repo as a model-checkpoint with interact.py, following error was obtained -

Command used - `python3 ./interact.py --model_checkpoint='./finetuned_chatbot_gpt'`

Error:
Traceback (most recent call last):
File "./interact.py", line 155, in
run()
File "./interact.py", line 128, in run
model = model_class.from_pretrained(args.model_checkpoint)
File "/usr/local/lib/python3.6/dist-packages/transformers/modeling_utils.py", line 558, in from_pretrained
model.__class__.__name__, "\n\t".join(error_msgs)
RuntimeError: Error(s) in loading state_dict for OpenAIGPTLMHeadModel:
size mismatch for transformer.tokens_embed.weight: copying a param with shape torch.Size([40483, 768]) from checkpoint, the shape in current model is torch.Size([40478, 768]).

While just running `python3 interact.py`, the result was fine.
Kindly see to this issue

Compare the model URLs in the README and utils.py, then trace how interact.py passes --model_checkpoint to model_class.from_pretrained. Reproduce the reported vocabulary-size mismatch and determine which checkpoint and tokenizer configuration are compatible; done means the documented model and the default script use consistent, runnable artifacts.

Written by the indexing model from the issue text.